fastLoess WebAssembly Out-of-Sample Prediction

Evaluate a fitted Batch model at query points that were not in the training set.

Overview

Retained-model prediction is available in Batch mode. Online also provides predict_window() for a fresh fit of its current bounded window; Streaming does not retain a complete training model for query-time prediction.

result.predict(newX, options) evaluates the fit at arbitrary query points, like R’s predict(model, newdata). Query points are flattened, dimensions values per point.

It always fits exactly, unlike fit()’s default surface_mode: "interpolation" — so predicting at a training point may not exactly match fit()’s output unless surface_mode: "direct" was used.

Requires retain_model: true on the constructor before fit(), otherwise predict() throws.

---

Options

Field	Type	Default	Description
outputs	string[]	[]	Optional fields: "se", "gradient" (or "derivative")
intervals	{ confidence?: number; prediction?: number }	disabled	Grouped confidence and prediction coverage levels.
extrapolation	string	"clamp"	Behavior for query points outside the training range, on any dimension
max_extrapolation_distance	number	disabled	Under "linear" extrapolation, the max allowed per-dimension distance beyond the training boundary before erroring
max_neighbor_distance	number	disabled	Max allowed distance to the farthest point in a query’s k-nearest-neighbor window before erroring

outputs

Select "se" to compute standard errors for each query point using the retained model’s residual scale and per-point leverage. Interval levels also enable standard errors. Select "gradient" (or "derivative") to include the local fit’s gradient (dimensions values per query point, flattened). Optional outputs are omitted by default.

intervals.confidence

Confidence level for the confidence interval around the mean response at each query point (e.g. 0.95). Uses the same z-score convention as fit()’s own confidence intervals. Disabled by default.

intervals.prediction

Confidence level for the prediction interval for a new observation at each query point (e.g. 0.95). Widens using the same residual scale fit() used for its own intervals when available, otherwise falling back to a MAD-based estimate. Disabled by default.

extrapolation

Behavior for query points outside the training range, on any dimension:

Policy	Behavior
"clamp" (default)	Clamps each out-of-range dimension to the nearest training boundary
"linear"	Linearly extrapolates from the boundary point’s local fit and gradient (first-order Taylor expansion)
"error"	Fails the whole call with an error

max_extrapolation_distance

Under "linear" extrapolation, the maximum allowed per-dimension distance beyond the training boundary before predict() throws, instead of returning an unbounded value. Disabled by default (uncapped).

max_neighbor_distance

Maximum allowed distance to the farthest point in a query’s k-nearest-neighbor window before predict() throws. Guards against an “empty corner” blind spot: a query point can sit inside every dimension’s min/max bounding box yet still be far from any real training data. Disabled by default (uncapped); measured as a plain (raw-coordinate) Euclidean distance, independent of distance_metric.

Example

Basic Usage

const { Loess } = require('fastloess-wasm');

const x = new Float64Array([1, 2, 3, 4, 5]);
const y = new Float64Array([2.1, 4.0, 6.2, 8.0, 10.1]);

const model = new Loess({ fraction: 0.7, retain_model: true });
const result = model.fit(x, y);

const prediction = result.predict(new Float64Array([1.5, 4.5]));
console.log("Predicted y:", prediction.y);

Predicted y: Float64Array(2) [ 3.05, 9.05 ]

Standard Errors and Derivative

const { Loess } = require('fastloess-wasm');

const x = new Float64Array([1, 2, 3, 4, 5]);
const y = new Float64Array([2.1, 4.0, 6.2, 8.0, 10.1]);

const model = new Loess({ fraction: 0.7, retain_model: true });
const result = model.fit(x, y);

const prediction = result.predict(new Float64Array([2.5]), {
    outputs: ["se"],
    outputs: ["derivative"],
});
console.log(prediction.y, prediction.standard_errors, prediction.derivative);

Float64Array(1) [ 5.1 ] undefined Float64Array(1) [ 2.2000000000000006 ]

Linear Extrapolation

const { Loess } = require('fastloess-wasm');

const x = new Float64Array([1, 2, 3, 4, 5]);
const y = new Float64Array([2.1, 4.0, 6.2, 8.0, 10.1]);

const model = new Loess({ fraction: 0.7, retain_model: true });
const result = model.fit(x, y);

const prediction = result.predict(new Float64Array([10.0]), {
    extrapolation: "linear",
});
console.log("Extrapolated y:", prediction.y);

Extrapolated y: Float64Array(1) [ 10.1 ]